Abstract: In order to sort the normal watermelon seeds, according to the characteristics of watermelon seeds, a feature extraction algorithm based on the gray scale was proposed, and its verification tests was carried out on a CCD color sorter. In the seeds image pre-processing, the contrast adaptive histogram equalization of the seeds image was executed. Then, after the median filter of the histogram equalization images, the value of the gray scale of the watermelon seeds was extracted as the classification characteristic quantity. The classification characteristic quantity was trained on the CCD color sorter, and the normal seeds were picked out finally with 95% recognition rate. The algorithm can provide theoretical support and technical realization for the classification and recognition of watermelon seeds.